We first outline some mathematical tools that are appropriate for describing traffic in optical networks. We use these tools to investigate OBS architectures, in two respects: the burst-creation process, and the multiplexing of bursts within a network. We find that OBS networks can achieve similar gains from statistical multiplexing as can packet-switched networks, unless switching overheads are significant, in which case OBS networks are more efficient. We weigh the tradeoff between fast-burst schemes like JET and slow-burst schemes like WROBS, and find that which is more efficient depends on the number of traffic flows per link.
We also use traffic traces to investigate the adaptability of OBS schemes. We argue that, while packet switched networks are inherently more adaptable than OBS networks, the latter are sufficiently adaptable to cope with high-volume traffic as seen in today's core links.